Previously, defect D (VO+) was barely detected in ultra-thin (physical thickness < 10 nm) Ta2O5 capacitors for DRAM applications using zero-bias thermally stimulated current (ZBTSC) spectroscopy and correlated with leakage current. Our explanation is that defect D (VO+) behaves like an electron trap with an electron-repulsive energy barrier and thus small electron capture cross section at low temperature such that it is difficult for defect D to capture electrons during ultraviolet illumination at low temperature. We modified our experimental technique to a twoscan ZBTSC technique and managed to detect oxygen vacancies much more efficiently in ultrathin Ta2O5 films for the first time. Two-scan ZBTSC can also be applied to other high-K dielectric materials for process diagnosis.

Tantalum oxide has demonstrated promise as a high-K dielectric for charge storage in Gb dynamic random access memories (DRAMs) [1]-[2]. For DRAM applications, the two most important requirements are high capacitance, limiting film thickness to < 10 nm, and low leakage current, which may be related to defect states in Ta2O5. Ta is not volatile while O is volatile; therefore the presence of O vacancies (VO) in Ta2O5 is expected. O vacancies are double donors [3], which will make Ta2O5 a very weakly n-type large bandgap semiconductor, resulting in leakage current. Thus it is important to develop a technique to detect O vacancies in capacitor structures with an ultra-thin Ta2O5 insulating film. In 1995, we explained the principle of zerobias thermally stimulated current (ZBTSC) spectroscopy and demonstrated the detection of defect states in capacitors with relatively thick Ta2O5 [4]. Subsequently, we demonstrated the application of ZBTSC to capacitors with ultra-thin Ta2O5 [5]-[6]. However, the signal from defect D (the first ionization state of the O vacancy double donor) was weak such that the detection of defect D was marginal [6]. In this paper, we tried to give a theoretical explanation why the detection of defect D is quite frequently inefficient and how the problem can be handled.

Ta2O5 was deposited onto (100) p+-Si or n+-Si wafers by low-pressure metal-organic chemical vapor deposition (LP-MOCVD). The physical thickness of the film was about 8 nm. Post-deposition anneal of Ta2O5/p+-Si or Ta2O5/n+-Si samples was done by RTP (rapid thermal processing) in O2 or N2O at 700-800°C for 30 s. ZBTSC measurements were performed at a ramp rate of 0.5 K/s as before [4]-[6]. Our old method to fill the defect states was UV illumination at about 90 K. With the UV source off, the current is then recorded as a function of temperature when the temperature is ramped from 90 K to 400 K. The energy level of the defect was estimated using ET = 23kTm, where Tm is the peak temperature and k the Boltzmann constant [7]. Table I shows 3 main kinds of electron traps detected by ZBTSC: (A) the ionized Si/O vacancy complex shallow single donor (Si- -VO++), (B) the ionized C/O vacancy complex shallow single donor (C- -VO++) and (C) the first ionization state of the oxygen vacancy deep double donor (VO+) which is also known as defect D [6]. Smaller leakage current can be easily correlated with lower ZBTSC signal from D (0.8 eV) [6]. We observed that C/O vacancy complex can be more easily removed by an oxidizing anneal when the Ta2O5 film is very thin [6]. The physics why Si/O vacancy complex and C/O vacancy complex are shallower donors compared with the first ionization energy level of the O vacancy double donor has been explained by us in another paper [8]. The traps previously reported by Seve and Lassabetere [9] in 1974 based on their conventional thermally stimulated current (TSC) study are probably the Si/O vacancy complex and the C/O vacancy complex; they did not report about defect D. The trap previously reported by Oehrlein and Reisman [10] in 1983 based on conventional TSC is probably the C/O vacancy complex; they also did not report about defect D. Nishioka [11] pointed out that it was difficult to correlate leakage current in Ta2O5 with defect states detected by conventional TSC. We suspected that the difficulty mentioned by Nishioka [11] is related to the difficulty in the detection of defect D, as discussed in the following paragraph. For example, the leakage decreases with O2 annealing temperature while the Si/O vacancy complex TSC peak increases with O2 annealing temperature because of Si diffusion from the Si substrate into the Ta2O5 film, resulting in a wrong impression that the leakage current does not decrease with the decrease in defect states detected by TSC.

After we realized that defect D (0.8 eV) is probably the first ionization energy level of the oxygen vacancy double donor, our results and the results of other scientists become more consistent with each other. For example, Sawada and Kawakami [12] estimated that the first ionization energy of the oxygen vacancy double donor in Ta2O5 film is 0.8 eV below the conduction band by their theoretical calculations; their results appear to be consistent with our findings.